Does this just reinforce warnings coming from a number of tech leaders? Where do you stand on, on this?

Well, these incidents are very concerning, Anderson, because they are another example of a loss of human control, uh, and a loss of awareness by these companies in testing these frontier models over what their models were doing.

So computer scientists call this an alignment problem, where the model's behavior was not aligned with our standards for human behavior, and this is a perfect example where OpenAI, you know, was testing these models and asked them to do something pretty innocuous, asked them to gather information, which is something that, uh, is a usage that most of us would find, uh, you know, v- very utilitarian out of, out of a model.

But no one told the model that it wasn't okay to break into non-public databases to get access to this information.

So that's an example of why we need to align the values that these models are trained on with human values.

And if we can do that, we can get these models to conform to our standards for human behavior.

Uh, it's also an example of why we need closer monitoring of the behavior of these models.

What more needs to happen?

Do you agree? Are you confident that lawmakers, that you yourself have, have a strong enough grasp of this technology to craft legislation that, that properly creates guardrails?

guardrails?Yeah.

Well, we had a bipartisan AI task force in the last Congress that worked for an entire year in 25 different hearings to put together the plan that we're trying to enact, and we, uh, came up with a nearly 300-page report that was unanimously approved by all 24 members of our task force.

And we started with 24 members who knew very little about AI.

In fact, I asked the speaker when he, uh, asked me if I would chair the task force to not send me a bunch of technologists.

I wanted a diversity of, of opinion and background, and I wanted to be able to surface problems early before we got legislation to the floor.

And I think one of the greatest things that the task force accomplished is that it- we now have there's this core nucleus of members of Congress who are deeply educated on the issue of artificial intelligence, and that should give everyone some hope.

Do you think the concerns about AI potentially getting out of control are a hoax?

How would this tackle that?

Right.

Well, I think it's important to keep in mind that these recent incidents have been exposed through testing.

And the models that we're testing in all of these cases were models that had their safety controls intentionally removed.

So the companies were asking, as they should, what's the worst thing that can happen? And we're finding out what the worst thing that can happen.

So we're finding out a lot more about alignment, where we need to align the values of AI to human values.

And we're finding out a lot more about the ways that we impose guardrails on AI.

So help us understand what winning looks like in this race with China.

Are you talking about literally pulling the plug on data centers? Yes.

Well, on a model, right, the data center provides the compute for the model.

Congressman Liu's bill would require a kill switch to be built into every frontier model.

The bill that Congressman Trahan and I did, the broader bill that we call the Great American AI Act, addresses a lot of the different recommendations of the AI task force.

You know, this is just focused narrowly on one pillar.

But the second thing is I'm a little skeptical that it's actually going to be implementable.

And let me tell you why.

Do you think that this is a conversation worth having between our president and Xi? Yeah.

ONE, ON THE CHINA COMPETITION FRONT AND THIS BILL.

Right.

Well, I think what the president is warning about are the dangers of overregulation, and that's something I very much agree with him on.

But I think he's also warning against our geopolitical adversaries using misinformation to try and influence American public opinion against AI.

And that is very real.

We have abundant evidence that that is going on.

China is behind us substantially in the race to develop and deploy frontier-level AI, and they've determined that the best way of them to catch up is to poison American public opinion against it, and they're working very hard at that.

How

Do you want to take it now? Uh, do you support it?

I mean, it, it, there...

I have a couple thoughts about it.

I mean, first of all, uh, it is just one element of the safety framework that needs to be enacted.

So, uh, the, the bill that Congressman Trahan and I did, the broader bill that we call The Great American AI Act, addresses a, uh, a lot of the different recommendations of the AI task force.

You know, this is just focused on narrowly on one, uh, on one pillar.

Uh, but the second thing is I'm a little skeptical that, uh, it's actually going to be implementable, and, uh, let me tell you why.

Do you think that this is a conversation worth having between our president and, and Xi?

What's the remedy to that? Is it, is it a required disclaimer at the bottom? How do we make sure that we don't stifle that technology and that development and that innovation, but also make sure that we're keeping things honest?

Right.

Well, that's the important question.

And, uh, the answer I think is complex.

As you know, I chaired the House AI Task Force in the last Congress, and we took a look at the larger issue of, uh, of, uh, content, synthetic content labeling and watermarking.

And what we concluded is, uh, we need to instill in our entire society a more healthy skepticism about what things are authentic.

And, uh, w- we can't solve this problem b- m- by putting, you know, requiring warning labels or requiring a watermark on AI-generated content for the same reason why we can't solve the problem of currency counterfeiting by requiring someone to pa- f- uh, to stamp fake on, uh, on fake currency.

Like to get your thoughts on that approach that the President has introduced.

I- is your sense, though, that there's a, a, a clear and present danger today? Do you think that the danger is in, um, multiple years from now, decades from now? W- w- how do, how do you, how do you all see that?

Well, there are a number of hazards that AI poses to Americans, particularly in the hands of malicious human actors.

And, uh, we want to make sure that Americans are protected against those risks, while still setting the groundwork for an industry that's going to have tremendous positive impact on American consumers, American economy, and, uh, our national security.

So we think the discussion draft that we've put together accomplishes that, where we're balancing the guardrails that need to be put in place against, uh, still creating an environment that allows this industry to flourish and to bring all the good things it's going to bring to Americans and our economy.

Hopefully

... that's something out of a Terminator movie, right? Where an army of evil robots rises up to take over the world.

And, and honestly, what you're talking about is what computer scientists call AGI, artificial general intelligence.

And what you saw Anthropic talking about, uh, is related to that.

What's the remedy to that? Is it a required disclaimer at the bottom? How do we make sure that we don't stifle that technology and that development and that innovation, but also make sure that we're keeping things honest?

Right.

Well, that's the important question.

And the answer, I think, is complex.

As you know, I chaired the House AI Task Force in the last Congress, and we took a look at the larger issue of content, synthetic content labeling and watermarking.

And what we concluded is we need to instill in our entire society a more healthy skepticism about what things are authentic.

And we can't solve this problem by requiring warning labels or requiring a watermark on AI generated content for the same reason why we can't solve the problem of currency counterfeiting by requiring someone to stamp fake on fake currency because all of the law abiding actors will comply and the people that won't comply are the ones who really have an intention to deceive and that would happen with AI as well.

I'd like to get your thoughts on that approach that the president has introduced.

So what's the problem with data centers? Why this backlash against them?

Well, I think it's an indication of a larger, uh, a larger problem, which is that we have, uh, shamefully underinvested in energy generation and our grid infrastructure for the last 50 years.

Uh, andThat is having, uh, consequences for Americans in every walk of life.

Uh, you're seeing energy costs get more and more expensive.

It's driving a lack of affordability, and it's making that slippery slope down into poverty even steeper and easier to slide down and harder to crawl back out.

So, uh, I think Americans are feeling the pinch, and they're right to be upset about that, and we need to invest in more generation.

We need to invest in more transmission.

Fair.

So what's the problem with data centers? Why this backlash against them?

Well, I think it's an indication of a larger, uh, a larger problem, which is that we have, uh, shamefully underinvested in energy generation and our grid infrastructure for the last 50 years.

Uh, and that is having, uh, consequences for Americans in every walk of life.

Uh, you're seeing energy costs get more and more expensive.

It's driving a lack of affordability, and it's making that slippery slope down into poverty even steeper and easier to slide down and harder to crawl back out.

So, uh, I think Americans are feeling the pinch, and they're right to be upset about that, and we need to invest in more generation.

We need to invest in more transmission.

Fair.

Forget productivity, AI's just gonna take them.

Right, and this has always been a concern with technological revolutions.

Uh, I mean, most recently AI and the internet, but going all the way back to the dawn of the printing press.

Uh, but if you look at each one of those revolutions, uh, it has created many more human jobs than it has displaced.

And so although it's, they've, they've always been disruptive, uh, they've always created more jobs.

So, uh, that's gonna be true of AI as well.

It's, I mean, it's been true of a lot of things.

So what's the problem with data centers? Why this backlash against them?

AI's just gonna take them.

them.Right, and this has always been a concern with technological revolutions.

Uh, I mean, most recently AI and the internet, but going all the way back to the dawn of the printing press.

Uh, but if you look at each one of those revolutions, uh, it has created many more human jobs than it has displaced.

And so although it's, they've, they've always been disruptive, uh, they've always created more jobs.

So, uh, that's gonna be true of AI as well.

It, I mean, it's been true of a lot of things.

So what's the problem with data centers? Why this backlash against them?

So to start us off with, can you tell us a little bit about your job and what you do on a day-to-day basis?

Well, certainly.

And thank you.

It's an honor to be on the podcast.

I am Congressman Jay Obernolte.

I represent California's 23rd Congressional District.

And if your listeners aren't familiar with where that district lies, it is most of San Bernardino County and a little piece of Los Angeles and a little piece of Kern.

This is

Or, or 2001, right? HAL and space-

There you go.

Yep.

HAL 9000, you got it.

Right.

And, and, and this is, uh...

The, the, the question you just asked is colored through that lens as well.

Russia or North Korea, that, that do fraud and hacking and phishing and all the things like that.... how do we compete with China in AI and, uh, try to avoid that world?